stream reasoning and linked data: a clash of...

Stream Reasoning and LinkedData: A Clash of Cultures

Andreas Harth

Karlsruhe Institute of Technology (KIT)

2nd RDF Stream Processing Workshop

ESWC 2017

Outline

• Act I: Status quo and life-changing event

• Act II: We have achieved some alignment but still no dice

• Act III: An attempt at convergence

ACT IStatus quo and life-changing event

„Nothing Endures But Change“

• Current and future scenarios around the Internet of Things and theWeb of Things require processing of (near) real-time data

• Frillions of sensors in the near future

• We need systems for processing live data (e.g., read a temperaturesensor or a website clickstream)

• The systems should also support actuators (e.g., set a heating/coolingsystem or send an email message)

Requirements

• Process data in near-realtime (sub-second) for live query, control, virtual reality/augmented reality

• Handle dynamics and behaviour (of both systems and humans)

• Connect different subsystems (sensors, actuators; tracking, simulation, rendering…)

• Integration of data sources and interoperation between systems

• Web-scale: simple, elegant, robust, decentralised, many participants

Which Data Model? Which Data Format?

RMI

Which Network Protocol?

Which IoT Platform?

IoT Hub

Semantic Web

• Semantic: Why not agree on RDF as data model?

• Web: Why not agree on HTTP as network protocol?

• Web: Why not agree on a decentralised architecture?

Web Architecture (1990)

https://www.w3.org/People/Berners-Lee/1996/ppf.html

Linked Data Architecture (2009)

https://www.w3.org/2009/Talks/0204-ted-tbl/#(7)

Web of Things Architecture (2015)

https://www.w3.org/2015/05/wot-framework.pdf

ACT IIWe have achieved some alignment but still no dice

Linked Data Interface

• Sensor has URI (e.g., http://localhost/sensor/temp)

• GET on http://localhost/sensor/temp returns current reading

SPARQL Query Current Temperature

SELECT ?temp

FROM <http://localhost/sensor/temp>

WHERE {

?x rdf:value ?temp .

}

Read-Write Linked Data Interface

• Actuator has URI (e.g., http://localhost/actuator/gpio1)

• PUT on http://localhost/actuator/gpio1 switches an electric consumeron or off (with the right RDF message body)

Simple Reflex Agents

Russel/Norvig. ArtificialIntelligence – A Modern Approach, Third Edition,Pearson Education, 2010.

Or other cognitive architectures: SOAR, ACT-R…

Von-Neumann Architecture (1945) von Neumann, John (1945), First Draft of a Report on the EDVAC.

Simple Thermostat Reflex Agent

• Loop forever:• GET http://localhost/sensor/temp• Query ?x rdf:value ?temp .• If ?temp > 25 °C:

• PUT http://localhost/actuator/heating• With some RDF to set heating to OFF

• If ?temp < 20 °C:• PUT http://localhost/actuator/heating• With some RDF to set heating to ON

• Wait 1 minute

• The Reflex Agent is acting as user agent

But I Want Streaming!

http://firehose.sandia.gov/images/firehose.jpg

Uniform Network Interface

• Components A and B, data flows from A (source/producer) to B (sink, consumer)

• REST assumes request/response communication pattern betweencomponents with client connector and server connector

• Clients emit requests, receive response

• Servers answer to incoming requests with a response

Network Interface: Push vs. Pull

Push

• A is client, B is server

• A emits PUT request

• At A: B.put(value)

• Loop at A

Pull

• A is server, B is client

• B emits GET request

• At B: value = A.get()

• Loop at B

Bi-directional Communication in HTTP

• Stream processor acts as server (which processes incoming HTTP requests)

• For publish/subscribe, each component is both client and server (a transducer)

Actor Model (1973)

“Actors communicate with each other by sending asynchronous messages. Those messages are stored in other actors' mailboxes until they're processed.”

http://www.brianstorti.com/the-actor-model/

Carl Hewitt; Peter Bishop; Richard Steiger (1973). "A Universal Modular Actor Formalism for Artificial Intelligence". IJCAI.

Message Bus (for Stream Processing)

http://www.enterpriseintegrationpatterns.com/img/MessageBusSolution.gif

Message bus systemscentralise the actors‘ mailboxes

Examples are:

• ROS

• Kafka

• …

Protocol/Architecture

Linked Data

• Distinction between client (useragent) and server

• User agents invoke requests, servers answer with responses

• Server is persistent

• No central message brokerneeded

Stream Processing

• No clear distinction betweenclient and server

• All components can send messages and receive messages

• All components are persistent

• Some variants rely on centralmessage broker

TimeThe New Media Reader, edited by Noah Wardrip-Fruin and Nick Montfort, book design by Michael Crumpton, 2003, http://www.newmediareader.com/, 50. Time Frames (from Understanding Comics), Scott McCloud, 1993

Aspect: State vs. Event

• Event: Leonard Cohen is born on September 21, 1934

[] a :Event ; :type :birth ; :date "1934-09-21" .

• State: Leonard Cohen

:lc :birthDate "1934-09-21" .

• Event: Leonard Cohen dies on November 7, 2016

[] a :Event ; :type :death ; :date "2016-11-07" .

• State: Leonard Cohen

:lc :birthDate "1934-09-21" .

:lc :deathDate "2016-11-07" .

https://www.youtube.com/watch?v=YD6fvzGIBfQ

Aspect of Data Representation

Linked Data

• Messages represent resourcestate

• Getting the combined currentstate of resources involves GETs on URIs

Stream Processing

• Messages represent events

• Some messages represent stateas well

• Getting the combined currentstate of resources requiresintegration of events

ACT IIIAn attempt at convergence

Selected Projects Using Reflex Agent Model

• ARVIDA: Applied Reference Architecture for Virtual Services and Applications• Break up monolithic industrial applications into distributed

applications• Components based on RESTful interfaces• RDF, RDFS and OWL for knowledge representation

(vocab.arvida.de)

• i-VISION: Immersive Semantics-based Virtual Environments for the Design and Validation of Human-centred Aircraft Cockpits • Human-Cockpit Operations Analysis• Semantic Virtual Cockpit• Virtual Cockpit Design Environment

Demo Setup

Demo Setup

Open Questions Stream Processing

• When to trigger query evaluation?• Every triple? (event-triggered)

• Every incoming request? (event-triggered)

• Every time slice? (time-triggered)

• When to trigger actions?

• How to integrate event and state messages?

• How to perform reasoning?• Over each message individually?

• Over entire state at each time step?

Conclusion

• In the Semantic Web community, building on HTTP and RDF seemsreasonable (to me)

• Linked Data and Stream Reasoning systems have different characteristics, leading to differences in user agent/server roles

• In dynamic environments, the modelling of aspect warrants someattention, especially in conjunction with request/response andreasoning

• Reflex agents provide a simple model for operating on resource state

• Linked Data notifications could provide a model for the handling ofevents in stream processing and stream reasoning systems

Linked Data Processing vs. Stream Processing

• What is the current temperature?

SELECT ?temp

FROM <http://localhost/sensor/temp>

WHERE {

?x rdf:value ?temp .

}

• Notify me when somebodyaccesses the web page foo.html!

tail -f access.log

| grep –e foo.html

| while read -r line; do

mail -s „hit“ harth@kit.edu;

done

UML State Diagrams vs. UML ActivityDiagrams

Light on

Light off

Switch on Switch off

Switch on

Switch off

Light on Light off

Boxes represent statesArrows represent events(transitions)

Boxes represent activitiesArrows represent states

Further Reading

• Andreas Harth, Tobias Käfer, Felix Leif Keppmann, Dimitri Rubinstein, René Schubotz, Christian Vogelgesang. "Flexible industrielle VT-Anwendungen auf Basis von Webtechnologien". VDE Kongress 2016, Internet der Dinge, Nov 7-8, 2016, Mannheim, Germany.

• Tobias Käfer, Sebastian Bader, Lars Heling, Raphael Manke and Andreas Harth. "Exposing Internet of Things Devices on REST and Linked Data Interfaces". 2nd International Workshop on Interoperability & Open Source Solutions for the Internet of Things. Co-located with 6th International Conference on the Internet of Things (IoT 2016). Nov 7, 2016, Stuttgart, Germany.

• Felix Leif Keppmann, Maria Maleshkova, Andreas Harth. "Semantic Technologies for Realising Decentralised Applications for the Web of Things". 21st International Conference on Engineering of Complex Computer Systems, Nov 6-8, 2016, Dubai, UAE.

• Sarah Brauns, Tobias Käfer, Dirk Koriath, Andreas Harth. "Individualisiertes Gruppentraining mitDatenbrillen für die Produktion". GI-Jahrestagung 2016.

• Andreas Harth, Tobias Käfer. "Towards Specification and Execution of Linked Systems". 28. GI-Workshop Grundlagen von Datenbanken, May 24 - 27, 2016, Nörten-Hardenberg, Germany.